查找表与神经网络相结合的图像逆半调算法

A Method of Inverse Halftoning Based on Lookup Tables and Neural Networks

为提高查找表逆半调算法中“未出现半调模式逆半调值”的估计精度,本文提出了基于IRN神经网络的逆半调逼近模型,通过分析、训练和优化确定了合适的IRN网络结构、隐层数、隐层节点数。实验结果表明,应用该算法训练、拟合出的查找表数据在逆半调重建图像的视觉效果及PSNR指标上表现良好,而且算法执行速度快、空间复杂度低。

To increase the estimation precision ot inverse halfroning values whose halftone patterns do not exist in the image training set, an inverse halftoning method based on lookup tables and the internal recurrent neural network is proposed. By studying, training and optimizing the appropriate neural network structure, layers and nodes are established. The proposed algorithm is computationally inexpensive and requires less memory than others. The experimental results show the satisfactory quality and higher PSNR.